Graphics Reference
In-Depth Information
Table 10.2
Dataset description
Set 1 Freeway
Set 2
Freeway with
lanes
vehicles
Set 3 Freeway
Set 4
Freeway
concrete
circular
surface
reflectors
Set 5 Urban road
with shadows
truth and the proposed technique is able to determine the lane features in “correct”
positions in the frame. The proposed technique is evaluated using the test video
datasets obtained by LISA-Q testbed [
10
]. The results are presented for five different
test image sequences that are listed in Table
10.2
, each dataset having a minimum of
250 image frames that are captured at 10-15 frames a second.
10.4.1 Accuracy Analysis
First, Fig.
10.5
shows some sample images with lanes that are extracted fromcomplex
road scenes by applying the proposed lane feature extraction method on input images
from the datasets listed in Table
10.2
. It can be seen that the proposed algorithm is able
to extract lanes in varying lane conditions, such as cracks (Fig.
10.5
a-d), presence
of vehicles (Fig.
10.5
e), presence of strong shadows (Fig.
10.5
e-h). The proposed
method is also able to extract lanes with circular reflectors as shown in Fig.
10.5
f, g.
Figure
10.6
shows detection accuracy results of the lanes in datasets 1, 2, and 3,
in which we are evaluating the detection of dashed lane markings (i.e., no circular
reflectors or solid lane boundaries). The effect of changing the number of scan bands
and the scan band width on detection accuracy is shown in Fig.
10.6
. It is evident that
reducing the number of scan bands will reduce the detection accuracy of the lane
features because depending on the position of the lane marker and the speed of the
vehicle, the scan band at a particular coordinate may fail to detect the lane marking
